检索结果-内蒙古大学图书馆

A Multi-Modal Assessment Framework for Comparison of Specialized Deep Learning and General-Purpose Large Language Models

引用

IEEE Transactions on Big Data 2025年第3期11卷 1001-1012页

作者： Nadeem, Mohammad Sohail, Shahab Saquib Madsen, Dag Oivind Alzahrani, Ahmed Abrahim Ser, Javier Del Muhammad, Khan Aligarh Muslim University Department of Computer Science Aligarh India VIT Bhopal University School of Computing Science and Engineering Bhopal-Indore Highway Kothrikalan Madhya Pradesh Sehore466114 India University of South-Eastern Norway Norway King Saud University Computer Science Department Community College Riyadh11437 Saudi Arabia TECNALIA Derio48160 Spain Department of Communications Engineering Bilbao48013 Spain Department of Applied Artificial Intelligence School of Convergence College of Computing and Informatics Seoul03063 Korea Republic of

Recent years have witnessed tremendous advancements in Al tools (e.g., ChatGPT, GPT 4, and Bard), driven by the growing power, reasoning, and efficiency of Large Language Models (LLMs). LLMs have been shown to excel in tasks ranging from poem writing and coding to essay generation and puzzle solving. Despite their proficiency in general queries, specialized tasks such as metaphor understanding and fake news detection often require finely tuned models, posing a comparison challenge with specialized Deep Learning (DL). We propose an assessment framework to compare task-specific intelligence with general-purpose LLMs on suicide and depression tendency identification. For this purpose, we trained two DL models on a suicide and depression detection dataset, followed by testing their performance on a test set. Afterward, the same test dataset is used to evaluate the performance of four LLMs (GPT 3.5, GPT 4, Google Bard, and MS Bing) using four classification metrics. The BERT-based DL model performed the best among all, with a testing accuracy of 94.61%, while GPT 4 was the runner-up with accuracy 92.5%. Results demonstrate that LLMs do not outperform the specialized DL models but are able to achieve comparable performance, making them a decent option for downstream tasks without specialized training. However, LLMs outperformed specialized models on the reduced dataset. © 2015 IEEE.

关键词： Contrastive Learning

来源：评论

学校读者我要写书评

暂无评论

Partition Strategies for Vertex-Centric Historical Graph Systems 25

Partition Strategies for Vertex-Centric Historical Graph Sys...

引用

Proceedings of the 40th ACM/SIGAPP Symposium on Applied Computing

作者： Alexandros Spitalas Georgios Tsolas Kostas Tsichlas Computer Engineering and Informatics Department University of Patras Rio Achaia Greece

ISBN: (纸本)9798400706295

This paper explores partitioning strategies for vertex-centric historical graph systems within distributed environments, focusing on efficient data management and query execution. Historical graphs, which capture the dynamic evolution of vertices and edges over time, present unique challenges for storage and computation due to their constantly changing nature. We investigate two partitioning approaches for offline and online environments. Both offline and online algorithms aim to minimize an appropriately defined notion of edge cuts within the historical graph setting; the former algorithm is based on BFS, while the latter is based on a greedy partitioning approach. Both algorithms are compared qualitatively and quantitatively to hash-based methods that are mainly used in this setting. They are evaluated on real-world datasets, using metrics such as weighted edge cut score ratio and load balance ratio. Our experiments reveal that our online method consistently reduces edge cuts with minimal overhead, making it suitable for real-time distributed processing systems for historical graphs. Similarly, the offline algorithm achieves much better results for weighted edge cuts, but it requires severely more computational resources to achieve this performance. Additionally, the study highlights how varying the number of workers and partitioning thresholds impacts system performance across different datasets. All our experiments have been carried out in a simulation environment. The results provide valuable insights into optimizing partition strategies for historical graphs, paving the way for more efficient graph analytics in distributed systems.

关键词： partitioning strategies

来源：评论

学校读者我要写书评

暂无评论

DR3DH: A DoS resistant extended triple Diffie–Hellman for mobile edge networks

引用

computer Networks 2025年 265卷

作者： Altaweel, Ala Bouridane, Ahmed Department of Computer Engineering College of Computing and Informatics University of Sharjah Sharjah United Arab Emirates

Mobile Edge Networks (MENs) are wireless networks that establish "close-to-end-users-clouds" to provide storage, computation, and communication services. MENs employ the Extended Triple Diffie–Hellman (X3DH) protocol to establish secret keys that provide end-to-end encryption among their nodes. X3DH relies on a cloud server that stores the public keys of the nodes to address the asynchronous scenarios when they operate offline. The cloud server in X3DH protocol represents a single point of failure (SPoF), as the exchange of secure keys is disrupted if the server fails or is compromised. Moreover, a malicious attacker, Eve, can launch various DoS attacks against the MEN nodes (e.g., packet dropping or packet modification) and wireless links (e.g., jamming) to disrupt the execution of X3DH during secure key exchange. This paper proposes the DoS-Resistant-X3DH (DR3DH) protocol to resolve the SPoF of the cloud server in X3DH and to address packet dropping, packet modification, and jamming DoS attacks. DR3DH employs Reed–Solomon Erasure Coding to encode the X3DH public keys into fragments that are distributed into a set of storage nodes. These storage nodes, which are determined by an Integer Linear Programming (ILP) solver, are optimum. That is, they have lowest probabilities to launch packet dropping/modification attacks and lowest probabilities that the wireless links toward them are jammed. The authors evaluated DR3DH through trace-driven simulations in MATLAB using two mobility traces and a proof-of-concept implementation in Java. The authors also compared DR3DH with two approaches that employ Random and Greedy policies when selecting the storage nodes for keys fragments. The results demonstrate the resistance of DR3DH to DoS attacks, achieving a high success rate that outperforms the Random and Greedy approaches by between 13.6% and 83.0%. Additionally, the results confirm the feasibility of DR3DH under different mobility models and node speeds, as demonstrated by

关键词： Packet loss

来源：评论

学校读者我要写书评

暂无评论

Multi-Normalization Residual Graph Convolutional Network for 3D Human Pose Estimation

Multi-Normalization Residual Graph Convolutional Network for...

引用

2025 International Workshop on Advanced Imaging Technology, IWAIT 2025

作者： Pramono, Andy Chang, I-Cheng Puspasari, Betty Dewi Animation Study Program Universitas Negeri Malang Jawa Timur Malang Indonesia Department of Computer Science and Information Engineering National Dong Hwa University Hualien Taiwan Department of Informatics Engineering Institut Teknologi Nasional Malang Jawa Timur Malang Indonesia

ISBN: (数字)9781510688131

ISBN: (纸本)9781510688124

3D human pose estimation (HPE) has improved significantly through Graph Convolutional Networks (GCNs), which effectively model body part ***, GCNs have limitations, including uniform feature transformations across nodes and reliance on skeleton-based graphs that may miss complex motion *** address these issues, we introduce a Multi-Normalization Residual Graph Convolutional Network that fine-tunes the graph structure through affinity multi-normalization and activation, allowing the representation of additional connections beyond the *** extensive ablation study shows that this approach enhances performance with minimal overhead while maintaining the same model size, consistently outperforming state-of-the-art techniques on two benchmark datasets. © 2024 SPIE.

关键词： Musculoskeletal system

来源：评论

学校读者我要写书评

暂无评论

Progressive Semantic-Visual Alignment and Refinement for Vision-Language Tracking

引用

IEEE Transactions on Circuits and Systems for Video Technology 2025年第5期35卷 4271-4286页

作者： Liang, Yanjie Wu, Qiangqiang Cheng, Lin Xia, Changqun Li, Jia Peng Cheng Laboratory Shenzhen518000 China City University of Hong Kong Department of Computer Science Hong Kong Xiamen University Fujian Key Laboratory of Sensing and Computing for Smart City School of Informatics Xiamen361005 China Beihang University State Key Laboratory of Virtual Reality Technology and Systems School of Computer Science and Engineering Beijing100191 China

In recent years, vision-language tracking has drawn emerging attention in the tracking field. The critical challenge for the task is to fuse semantic representations of language information and visual representations of vision information. For this purpose, several vision-language tracking methods perform early or late fusion to fuse visual and semantic features. However, these methods cannot take full advantage of the transformer architecture to excavate useful cross-modal context at various levels. To this end, we propose a new progressive joint vision-language transformer (PJVLT) to progressively align and refine visual embedding with semantic embedding for vision-language tracking. Specifically, to align visual signals with semantic signals, we propose to insert a semantic-aware instance encoder layer (SAIEL) into each intermediate layer of transformer encoder to perform progressive alignment of visual and semantic features. Furthermore, to highlight the multi-modal feature channels and patches corresponding to target objects, we propose a unified channel communication patch interaction layer (CCPIL), which is plugged into each intermediate layer of transformer encoder to progressively activate target-aware channels and patches of aligned multi-modal features for fine-grained tracking. In general, by progressively aligning and refining visual features with semantic features in the transformer encoder, our PJVLT can adaptively excavate well-aligned vision-language context at coarse-to-fine levels, therefore highlighting target objects at various levels for more discriminative tracking. Experiments on several tracking datasets show that the proposed PJVLT can achieve favorable performance in comparison with both conventional trackers and other vision-language trackers. © 1991-2012 IEEE.

关键词： Semantics

来源：评论

学校读者我要写书评

暂无评论

Identify spoofing attacks in Internet of Things (IoT) environments using machine learning algorithms

引用

Journal of High Speed Networks 2025年第1期31卷 61-70页

作者： Vajrobol, Vajratiya Saxena, Geetika Jain Pundir, Amit Singh, Sanjeev B. Gupta, Brij Gaurav, Akshat Rahaman, Mosiur Institute of Informatics and Communication University of Delhi New Delhi India Department of Electronics Maharaja Agrasen College University of Delhi New Delhi India Department of Computer Science and Information Engineering Asia University Taichung Taiwan Dehradun India UCRD Chandigarh University Chandigarh India Department of Computer Science Ronin Institute MontclairNJ United States International Center for AI and Cyber Security Research and Innovations Asia University Taichung Taiwan

With the growing adoption of Internet of Things (IoT) devices, security concerns are becoming increasingly urgent. Protecting IoT systems from cyberattacks is crucial to safeguard sensitive information. Spoofing, particularly Domain Name System (DNS) and Address Resolution Protocol (ARP) spoofing, is a type of attack that can manipulate network traffic and compromise data integrity. DNS spoofing redirects users to fraudulent websites by altering domain name resolutions, while ARP spoofing tricks the network by associating a legitimate internet protocol address with a malicious MAC address, allowing attackers to intercept or modify communication. This study aims to develop an efficient method for detecting these types of spoofing attacks in IoT environments using machine learning techniques. The results show that the random forest algorithm outperforms other models, achieving remarkable performance with a 95.1% accuracy, a precision score of 95.2%, and a strong F1 score of 95.1%. A key contribution of this research is the simultaneous detection of both DNS and ARP spoofing within a unified framework, utilizing a comprehensive set of 46 features. These findings underscore the importance of ensuring robust protection against spoofing attacks to maintain the security and integrity of IoT systems. © The Author(s) 2025.

关键词： Adversarial machine learning

来源：评论

学校读者我要写书评

暂无评论

FC Portugal: RoboCup 2024 3D Simulation League Champions

FC Portugal: RoboCup 2024 3D Simulation League Champions

引用

27th RoboCup International Symposium, 2024

作者： Abreu, Miguel Mota, Pedro Azevedo, Tomás Silva, Francisco Reis, Luís Paulo Lau, Nuno Florido, Mário LIACC/LASI/FEUP Artificial Intelligence and Computer Science Lab Faculty of Engineering of the University of Port University of Porto Porto Portugal LIACC/LASI/FCUP Artificial Intelligence and Computer Science Lab Faculty of Sciences of the University of Porto University of Porto Porto Portugal IEETA/LASI/DETI Institute of Electronics and Informatics Engineering of Aveiro Department of Electronics Telecommunications and Informatics University of Aveiro Aveiro Portugal

ISBN: (纸本)9783031858581

FC Portugal, a simulated humanoid soccer team from the universities of Porto and Aveiro, achieved the third consecutive victory in the main competition of the 2024 RoboCup 3D Simulation League. The team registered 16 wins, 3 ties, and 2 losses, scoring a total of 105 goals while only conceding 17. FC Portugal also ended first in two secondary events: the Scientific Challenge, which evaluates innovative scientific contributions within the league, and the Fat Proxy Challenge, where teams share a common set of low-level skills and compete for the best high-level strategy. The codebase used in this competition was developed in Python after RoboCup 2021. This paper presents a brief overview of its structure along with the new features developed for this year’s competition. The key developments include a hierarchical learning framework, presented at the Scientific Challenge, and the release of our codebase to the RoboCup community. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.

关键词： 3D modeling

来源：评论

学校读者我要写书评

暂无评论

Public Roads Management and Visualisation

Public Roads Management and Visualisation

引用

International Symposium on Applied Machine Intelligence and informatics (SAMI)

作者： Jaroslav Marko Anton Baláź Department of Computer and Informatics Faculty of Electrical Engineering and Informatics Technical University of Košice Košice Slovakia

ISBN: (数字)9798350379365

ISBN: (纸本)9798350379372

The paper deals with the issue of collecting and displaying geographic information about roads in Slovakia with emphasis on information about obstacles on roads. The main objective is to design a solution for collecting and displaying the roads based on the analysis of existing systems and identifying the requirements for an optimal design. The paper starts with an analysis of available solutions and the development of requirements for the most favourable design, followed by the planning of an optimal solution for collecting and displaying road obstruction data. The focus of the paper is on the development of a method for creating and modifying geographic obstacle information, the design of a textual and graphical display of the information, and the design of a three-level model for user access to the application. The paper concludes with a discussion of the proposed solution, comparing the requirements for the optimal design with the developed design.

关键词： Roads Planning Object recognition informatics Machine intelligence

来源：评论

学校读者我要写书评

暂无评论

The Novel DeepBoost Multi Output Regressor for Household Finance Intelligence

The Novel DeepBoost Multi Output Regressor for Household Fin...

引用

Advanced Research in Computing (ICARC), International Conference on

作者： S. H. S. V. Suwandaratna G. O. Priyankara Department of Computer Science and Engineering Informatics Institute of Technology Colombo Sri Lanka

ISBN: (数字)9798331530983

ISBN: (纸本)9798331530990

This paper forecasts the microeconomic level household expenditures using a novel hybrid deep learning approach. In terms of research significance, household finance control has a major influence on the finance system within the economy. Accurate forecasting of household finances assists in maintaining positive financial behavior among individuals and the economy. The DeepBoost multi-output regressor proposed in this paper is based on the 1D CNN-ANN and the XGBoost. The proposed model in this paper is compared with the R 2 , MSE, and MAE since it’s a regression problem. The experimental results reveal that the proposed DeepBoost multi-output regressor has the best application in forecasting the multiple expenditures of households by outperforming the ANN, 1D CNN-ANN, and Random Forest Regressor models. The proposed DeepBoost multi-output regressor evaluated the housing, food, transportation, healthcare, other necessities, childcare, and tax expenditures that had 0.94, 0.98, 0.83, 0.94, 0.97, 0.97, and 0.99 values for the R 2 , 9037.71, 2692.12, 9788, 15077.33, 1373.93, 13629.36, and 1904.52 values for the MSE, and 66.07, 34.05, 73.17, 87.05, 26.25, 78.74, and 29.47 MAE values than the ANN, RFR, and 1D CNN-ANN models.

关键词： Deep learning Pediatrics Computational modeling Finance Transportation Medical services Predictive models Microeconomics Forecasting Random forests

来源：评论

学校读者我要写书评

暂无评论

Compact and Efficient Neural Networks for Image Recognition Based on Learned 2D Separable Transform

Compact and Efficient Neural Networks for Image Recognition ...

引用

Digital Signal Processing and its Applications (DSPA), International Conference on

作者： Maxim Vashkevich Egor Krivalcevich Department of Computer Engineering Belarusian State University of Informatics and Radioelectronics Minsk Belarus

ISBN: (数字)9798331518752

ISBN: (纸本)9798331518769

The paper presents a learned two-dimensional separable transform (LST) that can be considered as a new type of computational layer for constructing neural network (NN) architecture for image recognition tasks. The LST based on the idea of sharing the weights of one fullyconnected (FC) layer to process all rows of an image. After that, a second shared FC layer is used to process all columns of image representation obtained from the first layer. The use of LST layers in a NN architecture significantly reduces the number of model parameters compared to models that use stacked FC layers. We show that a NN-classifier based on a single LST layer followed by an FC layer achieves 98.02% accuracy on the MNIST dataset, while having only 9.5k parameters. We also implemented a LST-based classifier for handwritten digit recognition on the FPGA platform to demonstrate the efficiency of the suggested approach for designing a compact and high-performance implementation of NN models.

关键词： Accuracy Image recognition Quantization (signal) Two-dimensional displays Transforms Artificial neural networks computer architecture Image representation Field programmable gate arrays Image classification

来源：评论

学校读者我要写书评

暂无评论

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案：

请选择收藏分类：

通借通还

建议与咨询 留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案： 新增检索档案 确定 取消

请选择收藏分类： 新增自定义分类 确定 取消

通借通还

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

请选择保存的检索档案：

请选择收藏分类：